Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Assessing the uncertainty of soil-moisture impact on convective
precipitation by an ensemble approach
Olga Henneberg1 , Felix Ament2 , and Verena Grützun2
1
2
Institute for Atmospheric and Climate Science, ETH Zurich
Meteorological Institute, Uni Hamburg
Correspondence to: Olga Henneberg ([email protected])
Abstract. Soil moisture influences the occurrence of convective precipitation. Therefore an accurate knowledge of soil moisture
might be useful for an improved prediction of convective cells. But still the model uncertainty overshadows the impact of soil
moisture in realistic cases even in 1 km resolution and therefore convection resolving models. Only drastic soil moisture
changes can exhibit the model uncertainties but the systematic behaviour is still complex and depends strongly on the strength
5
of soil moisture change.
Here we performed seven experiments with modified soil moisture using an ensemble approach for each experiment. Only a
50% soil moisture enhancement and a complete dried soil impact precipitation patterns considerably in structure, amplitude
and location in certain analysis areas. Both, the enhanced and reduced soil moisture result in a reduced precipitation rate.
Replacing the soil moisture by a realistic field from different days influences the precipitation insignificantly. We point out the
10
need for uncertainty estimations in soil moisture studies.
1
Introduction
Convective precipitation changes rapidly in time and is very variable in space (Pedersen et al., 2010). The heterogeneity of
convective precipitation and the interaction of different scales challenge atmospheric models on the global and regional scale.
Nowadays regional climate models operate with a 1 km scale resolution to represent convective processes explicitly and to
15
improve weather forecast (Mass et al., 2002). Nevertheless, precipitation formation undergoes a complex chain of atmospheric
processes from the micro to the synoptic scale (Richard et al., 2007). Therefore precipitation remains a highly uncertain quantity. The final precipitation formation includes unresolved microphysical conversion processes, most often including the ice
phase (Field and Heymsfield, 2015), that rely on complex parametrisations introducing a large uncertainty in the model. Many
studies found buffering effects for those processes (Muhlbauer et al., 2010; Glassmeier and Lohmann, 2016). Such a modifying
20
effect does not exist for the dynamical influence on convective precipitation, such as baroclinic and moist conditional instability. Soil moisture stands at the beginning of the convective precipitation formation that is highly sensitive to the aforementioned
atmospheric stratification.
Soil moisture affects the partitioning of turbulent heat fluxes into sensible and latent heat, which once affects the surface temperature due to the latent heating. The surface temperature plays a crucial role in the initiation of convection. Second the soil
1
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
moisture strongly influences the specific water content via latent heat flux. Furthermore, the specific water content in the lower
troposphere modifies moist conditional instability. On the one hand high surface temperatures can be reached and initiate convection with a low soil moisture content. On the other hand, high soil moisture can destabilise the atmosphere by introducing
water vapour in the lower troposphere favouring convection as well. These competing effects hamper the analysis on soil mois5
ture influence. Many parameters to describe the atmospheric stability react on the soil moisture. A strong systematic effect on
soil moisture changes exists for the latent and sensible heat fluxes, as well as equivalent potential temperature, lifting condensation level and convective energy, that following the process chain (Barthlott et al., 2011). Despite to the systematic effect
on partitioning of the heat fluxes, precipitation reacts less systematically on soil moisture variations (Barthlott and Kalthoff,
2011; Hohenegger et al., 2009). The distribution and inhomogeneity of soil moisture patterns may initiate secondary circulation
10
(Clark et al., 2004; Adler et al., 2011; Kang and Bryan, 2011; Dixon et al., 2013; Maronga and Raasch, 2013; Froidevaux et al.,
2014).
Accordingly, there is no clear consent on soil moisture precipitation interaction in literature: Barthlott et al. (2011) found a
strong dependency of precipitation with changes larger than 500% for a soil moisture variation of ±25% in regions with low
15
mountain ranges and changes up to −75% for domains with higher mountain ranges. Significant differences between planetary
boundary driven and synoptic forced conditions could not be detected. Further studies by Kalthoff et al. (2011) and Hauck et al.
(2011) over orographic complex terrain investigate the role of orographic effects in the synergy of soil moisture-precipitation
feedbacks. Hauck et al. (2011) determines large systematic differences between modelled and observed soil moisture. The
influences on simulated precipitation is more complex and depends strongly on the chosen case and domain. A dependency of
all convective indices on the equivalent potential temperature was found by Kalthoff et al. (2011) over different orographic ter-
20
rains. However, convection was predominantly initiated over mountain, independent of the instability indices, but with smaller
convective inhibition. The dependency of equivalent potential temperature on soil moisture was influenced by surface inhomogeneity. Barthlott and Kalthoff (2011) provide a sensitivity study, in which the soil moisture was increased by ± 50% in steps
of 5%. While a systematic effect on the 24 hours precipitation sum for reduced soil moisture exists, precipitation does not react
systematically in wetter simulations.
25
Diversity in the results may partly be attributed to model uncertainty. Hohenegger and Schär (2007) investigated the error
growth of random perturbation-methods in cloud-resolving models using time shifted model simulations and perturbed temperature fields in the initial conditions. In their model study with a resolution of 2.2 km rapid error growth was found far away
from perturbed regions, but growth of uncertainties is limited by the large-scale atmospheric environment. A further aspect of
model uncertainties is provided by the model resolution especially in terms of convection. Different results of soil moisture-
30
precipitation feedback occur for simulations with explicit and differently parametrised convection (Hohenegger et al., 2009).
Hohenegger et al. (2008) found different results in sign and strength of the influence of soil moisture depending on the used
model resolution. Simulations with explicitly resolved convection indicate a negative soil moisture-precipitation feedback, that
is in consent with many other studies as Barthlott and Kalthoff (2011) summarised.
As Richard et al. (2007) already states, convective precipitation suffers strongly from model uncertainty such caused by ini-
35
tial and boundary data. This study provides an uncertainty estimation that enables to distinguish between random changes in
2
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Figure 1. Complete model domain over Northern Germany for CTRL run. The two analysis areas are marked with red and blue rectangles.
precipitation and changes that results from differences in soil moisture. With this uncertainty estimation that is based on many
simulations with slightly different model set-ups the effect of different soil moisture modifications on precipitation can be
ranged and separated from random effects.
5
2
Soil moisture perturbation and its influence on precipitation
We simulate the convective introduced precipitation, observed on 03 August 2012, using the non-hydrostatic model COSMO
(Schättler et al., 2009) with a resolution of 1 km over Northern Germany including 400 x 450 grid points (Fig. 1) with various
simulation set-ups. A 1 km resolution allows an explicit representation of convection and provide much more accurate simulation of convective precipitation (Leutwyler et al., 2016, and references therein). Land surface processes are calculated by the
10
interactive soil and vegetation model TERRA-ML and coupled to atmospheric processes (Doms et al., 2011). Boundary and
initial conditions are provided by the coarse grid COSMO operational analysis with a resolution of 2.8 km.
A series of simulations include various soil moisture modifications of different strength and different realisations (table 1).
Two classes of changes to the soil moisture field are applied: extreme artificial changes, that show the full range of soil moisture
influence and realistic modifications (Fig. 2). Among the strong modifications are total drying of soil (Fig. 2c) and enhancement
15
of 50% (Fig. 2d). Those changes are applied once over the whole model domain and secondly over the red framed domain in
Fig. 1). A further artificial modification is redistribution into four alternating bands with 50% enhanced respectively reduced
soil moisture (Fig. 2b). Realistic but less intense modifications are implemented by replacing the soil moisture patterns by
those from another day (Fig. 2e and f). Therefore the soil moisture field from 20 August 2012 is used. On that day the soil
moisture content in the uppermost layer (1.2 mm [H2O] averaged over all land points in the uppermost 10 mm) is 0.3 mm
3
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Figure 2. a) soil moisture for CTRL run and differences between CTRL run and b) run BAND, c) run DRYa , d) run MOIa ,e) run REAL0820
and f) run REAL0719 in the uppermost soil layer.
lower than on 03 August 2012. On 19 July 2012 soil moisture content was high (1.9 mm [H2O]) and therefore this day was
used to simulate 03 August 2012 with realistic but higher soil moisture. The high uncertainty of convective precipitation on the
initial and boundary data (Richard et al., 2007) is accounted by an ensemble approach conducting additional simulations with
shifted boundaries by ten to 30 grid points. Those simulation will be explained in detail in Sect. 3. Here we will focus on the
5
simulation with shifted domain by ten grid points first.
In comparison between the CTRL run (Fig. 3a) and the simulation with shifted boundaries (Fig. 3b) differences in the single
cells in the West, partly over the North Sea, and in the structure of the large precipitation pattern in the East become obvi4
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Table 1. Model simulations with modified soil moisture (SM). Simulations are named by the applied soil moisture modification and with a
for whole model domain and p for modification in a subdomain (partly). Simulations with additional random changes are denoted with ii
and jj what represents the shifting of the model domain in grid points (For details see table 2).
simulation
Characteristics
modification
additional random changes
area
CTRL
LOCii jj
DRYa
dry out
whole model domain
DRYa ii jj
DRYp
dry out
area “red”
DRYp ii jj
MOIa
50% increased SM
whole model domain
MOIa ii jj
MOIp
50% increased SM
area “red”
MOIp ii jj
BAND
four bands
whole model domain
BANDii jj
REAL0820
SM from 20.08.12
whole model domain
REAL0820 ii jj
REAL0719
SM from 19.07.12
whole model domain
REAL0719 ii jj
TIMEtt
ous. These differences are predicated to the shifted boundary conditions by ten grid points (10 km). The brutal changes in
soil moisture cause even more obvious changes in the precipitation patterns. The enhancement of soil moisture in either the
whole domain or a sub domain changes the location of the precipitation for the chosen time dramatically (Fig. 3e and f). In the
moist simulation precipitation occurs mainly at places, that are free of precipitation in the CTRL run, and vice verse. Moder5
ate changes in soil moisture, such as applied by using realistic moisture fields, result in smaller changes in precipitation. The
general pattern observed in the CTRL run remains in REAL0820 and REAL0719 (Fig. 3f and g).
Figure 3 shows results of a single output time step only, but gives evidence that random perturbations in the simulations may
influence precipitation in a similar order of magnitude as effects due to soil moisture modifications. Detailed and statistically
reliable results for an extended estimation of the model uncertainty by a sufficient number of simulation and an analysis method
10
over all time steps is required. Both methods will be introduce in the following section.
3
Estimation of model uncertainties
The comparison of precipitation patterns between 10:00 UTC and 18:00 UTC for multiple simulations provide representative
result by using the SAL-score (Wernli et al., 2008) for every single time step. The SAL-score calculates a rate for the differences
in structure S, amplitude A and location L of precipitation patterns.
15
Amplitude A yields the differences of precipitation amount over the whole analysed domain:
A=
D(Rmod ) − D(Rcomp )
0.5[D(Rmod ) + D(Rcomp )]
(1)
5
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Figure 3. Precipitation rate at 14:45 UTC for a) CTRL run , b) LOC 10 00 and c - f) different soil moisture modified simulations.
with D(R) the averaged precipitation amount for modified model simulation denoted with mod and the compared simulation
that is mostly the CTRL run denoted with comp:
D(R) =
1 X
Rij
NGP
(2)
(i,j)∈
6
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
with the precipitation rate Rij within a grid point that is given by the indices i, j and the number of all grid points NGP in
the analysed domain. Component location L compares the location of precipitation in the two model simulations in two steps.
First the normalised distance of the centres of mass x(R) of the precipitation patterns in each model simulation is calculated.
L1 =
5
|x(Rmod ) − x(Rcomp )|
,
d
(3)
where d denotes the maximal distance within the analysed domain.
Secondly the distances from the centre of mass of all M individual cells xn to the centre of mass fore the whole precipitation
field x is calculated
PM
Rn |x − xn |
r(R) = n=1
PM
n=1 Rn
10
(4)
and compared:
|r(Rmod ) − r(Rcomp )|
L2 = 2
.
d
(5)
Both components of L are added.
Structure component S gives a hint whether the precipitation patterns tend to more convective precipitation with small but more
peaked rain objects or shallow precipitation with larger but less precipitating objects. Therefore a volume V (R) is calculated
15
by the sum of precipitation Rij over all grid cells within a precipitation cell n and the maximal precipitation Rnmax within
this cell:
Vn =
X
(i,j)∈
Rij
,
Rnmax
(6)
PM
n=1 Rn Vn
V (R) = P
.
M
n=1 Rn
(7)
With V (R) the volume over all precipitation cells M the structure component can be calculated similar to Eq. (1):
20
S=
V (Rmod ) − V (Rcomp )
.
0.5[V (Rmod ) + V (Rcomp )]
(8)
For more detailed information on SAL see Wernli et al. (2008).
To address the dependency of SAL-score on the chosen analysis area two different analysis areas are chosen (Fig. 1). The blue
framed area in Fig. 1 includes mainly the small convective cells and the red framed area includes the whole precipitation field.
For those two analysis areas two simulations are compared to each other respectively.
25
The significance of soil moisture impact is proven by facing with uncertainty estimations. Random perturbations are introduced
by shifting the domain boundaries by ten to 30 grid points north- and eastwards (table 2). Those perturbations provide an
estimation of the uncertainty caused by the chaotic behaviour of the atmospheric system and are superimposed on all systematic
and physical changes caused by the soil moisture perturbations. This method conserves the structure of all meteorological
7
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Table 2. Uncertainty-ensemble with randomly changed model simulations by model domain shifting, denoted with LOC and the number of
shifted grid points, and by shifting the model start time denoted with TIME. The shifted time is given in hours. The lower left corner of the
simulation domains is given in geographical (rotated) coordinates with the north pole being shifted to 40◦ N and -170◦ E.
run
corner in ◦ N
corner in ◦ E
starttime (UTC)
CTRL
50.87 (1.0)
15.55 (3.5)
0:00
LOC 00 10
50.97 (1.1)
15.56 (3.5)
0:00
LOC 00 20
51.07 (1.2)
15.57 (3.5)
0:00
LOC 00 30
51.17 (1.3)
15.59 (3.5)
0:00
LOC 10 00
50.88 (1.0)
15.39 (3.4)
0:00
LOC 10 10
50.98 (1.1)
15.40 (3.4)
0:00
LOC 10 20
51.08 (1.2)
15.42 (3.4)
0:00
LOC 20 00
50.89 (1.0)
15.23 (3.3)
0:00
LOC 20 10
50.98 (1.1)
15.25 (3.3)
0:00
LOC 20 20
51.08 (1.2)
15.26 (3.3)
0:00
LOC 30 00
50.89 (1.0)
15.08 (3.2)
0:00
TIME 01
50.87 (1.0)
15.55 (3.5)
1:00
TIME 02
50.87 (1.0)
15.55 (3.5)
2:00
TIME 03
50.87 (1.0)
15.55 (3.5)
3:00
TIME 04
50.87 (1.0)
15.55 (3.5)
4:00
TIME 05
50.87 (1.0)
15.55 (3.5)
5:00
TIME 06
50.87 (1.0)
15.55 (3.5)
6:00
input fields and does not create errors on a scale that can interact with the analysed processes. Furthermore, shifted start time
of the simulations (Hohenegger and Schär, 2007) provide an additional uncertainty. A time shift of one to six hours is also
applied to the CTRL run to extend the uncertainty estimation. The ensemble, further called the uncertainty-ensemble, delivers
17 independent model simulations including the CTRL run to estimate the uncertainty. Using the SAL-score an uncertainty
5
estimation with the uncertainty-ensemble will be done by comparing the simulations within the uncertainty-ensemble to each
other. Therefore all simulations with shifted model domain are compared to every other simulation with shifted model domain.
Those with shifted model start are compared only to the CTRL run. Positive or negative amplitude arises just as an effect of
which simulation is used as the CTRL run. To avoid an uncertainty range tending more to one direction, all comparisons are
done also in the reversed direction. This approach provides a symmetric distribution for the deviation range. The uncertainty
10
estimation (Fig. 4) encompasses a sample of 122 (number of simulations) times 32 (time steps) values even though, not all of
them are independent from each other.
Negative deviations in amplitude are mostly connected to negative deviations in structure and vice versa. Hence a reduction in
8
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Figure 4. SAL results for the uncertainty-ensemble between 10:00 UTC and 18:00 UTC. Structure is represented on the x-axis, amplitude on
y-axis and location by marker colours. Every marker presents a comparison between two model simulations at a single time step. Simulations
with shifted model domain are presented by filled dots and those with shifted model start time by rectangles. The borders of the grey rectangle
are calculated by 5% and 95% percentile of structure and amplitude.
rain’s amplitude arises by too small but peaked rain objects, whereas an increase in precipitation goes along with larger and
shallow rain objects. Largest deviations arise in the first hour until 11:30 UTC of the analysis time (Fig. 5). This accords with
the beginning of the precipitation event in the different simulations. The onset of precipitation differs in all simulations (not
shown) and therefore causes the largest uncertainties. The end of the precipitation event is not considered in the chosen time
5
range. A large shift in model start time leads to higher uncertainties. The random changes due to the shifted model domain do
not dependent on how much the boundaries are shifted. Conclusively, there are no systematics for locally perturbed simulations
but for time shifted simulations.
We define the model uncertainty for this study as the range from 5% percentile to 95% percentile for structure and amplitude
10
and by 90% percentile for location. Concerning to this definition, the uncertainty range is ±0.77 (±0.86) in structure, ±0.54
(±0.69) in amplitude, and up to 0.20 (0.29) for analysis area “red” (“blue”). Changes are defined as significant when the
response to the soil moisture modification is larger than the generated background noise. A residual probability of 10% remains
that the latter are not a result of soil moisture modification.
4
Significant effects of soil moisture modification on precipitation
Significant effects from soil moisture perturbations will be carved out with a comprehensive set of model simulations. All
15
simulations with modified soil moisture are conducted additionally with shifted domain boundaries as already applied to the
CTRL run (uncertainty-ensemble, see table 1). This huge amount of model simulations (a complete ensemble for each soil
9
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Figure 5. Amplitude values from Fig. 4 for comparisons to the CTRL run only on the single time steps.
moisture modification) and the uncertainty estimation from Sect. 3 accounts for a quantitative evaluation of the significance of
soil moisture influence on precipitation.
For all soil modified ensembles every ensemble member will be compared to the corresponding ensemble member (same shifting) from the uncertainty-ensemble. That delivers again a huge sample of SAL values (Fig. 6). Within every ensemble the
5
values are divided into those that exceed the uncertainty range (blue transparent filled rectangle in Fig. 6), that is given by
the uncertainty-ensemble, and those that are within. The percentage of the uncertainty range exceeding values is calculated to
decide whether a soil moisture modification leads to significant changes in precipitation (table 3). Changes caused by a soil
moisture modification will be treated as significant if more than 10% of the values exceed the uncertainty range. The threshold
is set to 10% because the uncertainty range was calculated by the 5% and 95% what remains a 10% probability that an exceed-
10
ing value can still be caused by model uncertainties.
The structural change of precipitation on soil moisture modification as in DRYp exceeds the uncertainty in only 5% (Fig. 6a
and table 3). For both scores S and A the percentage of exceeding values lies beneath the 10% threshold. Therefore, precipitation does not respond significantly on DRYp modifications, but for L in area “red” only. In contrast, the soil moisture reduction
in the whole domain affects the precipitation significantly (Fig. 6b). More than 50% of A exceed the uncertainty range and
15
some of them exceed it by far. For S only 11% of the values exceed the range. Nevertheless, this is enough to be treated as a
significant impact.
The soil moisture enhancement in a sub domain only, already results in significant precipitation changes, contrary to the drying
in the sub domain (Fig. 6c). Again the modification over the whole domain results in stronger response in precipitation.
The redistribution of soil moisture does not lead to any significant effects (Fig. 6e) but in location in area “blue”. The redistri-
20
bution of soil moisture increases the large-area heterogeneity, but decreases the small-area heterogeneity. This is in accordance
10
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Figure 6. SAL scatter plot: Comparison of ensembles a) DRYp , b) DRYa , c) MOIp , d) MOIa , e) BAND and f) MOI0820 with the uncertaintyensemble for area “blue”.
with Adler et al. (2011); Kang and Bryan (2011), who found an influence of redistribution of soil moisture on the location of
convective inition. Therefore, area “blue”, mainly containing small convective cells, is more influenced than area “red” with
the large advected precipitation band.
Even slight modifications of soil moisture, as Klüpfel et al. (2011) did by using different initialisation for soil moisture, lead
11
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Table 3. Percentages (pS , pA , pL ) of values S, A and L that exceed the model uncertainty by 90%. Uncertainties are in a range from
[−0.767, 0.767]([−0.857, 0.857]) in structure, [−0.538, 0.538]([−0.690, 0.690]) in amplitude and 0.200(0.288) in location for analysis
area “red” (“blue”). Bold values exceed model uncertainties in more than 10%. Averaged values and its deviation (S ± σ̂ 2 , A ± σ̂ 2 ). Bold
values are those mean values that differ significantly from the mean of the uncertainty-ensembleafter after Eq. (9) for an confidence interval
of 90%.
Structure
Ensemble
pS
Amplitude
S ± σ̂ 2
pA
analysis area “red”
DRYp
5.79
DRYa
23.14
MOIp
15.98
MOIa
9.92
BAND
3.03
REAL0820
3.31
REAL0719
2.48
3.58
0.02 ± 0.0034
0.30 ± 0.0063
22.31
−0.10 ± 0.0043
23.42
−0.03 ± 0.0022
0.28
−0.12 ± 0.0051
18.73
−0.04 ± 0.0025
0.55
0.05 ± 0.0024
1.65
−0.10 ± 0.0033
9.39
analysis area “blue”
DRYp
4.85
DRYa
11.82
MOIp
14.85
MOIa
15.76
BAND
7.27
REAL0820
0.91
REAL0719
1.21
0.08 ± 0.0058
51.21
−0.27 ± 0.0058
19.70
−0.04 ± 0.0026
0.30
−0.19 ± 0.0053
12.12
−0.10 ± 0.0045
0.91
−0.01 ± 0.0026
0.30
Location
A ± σ̂ 2
pL
−0.13 ± 0.0011
25.34
−0.05 ± 0.0042
8.26
−0.26 ± 0.0023
53.72
−0.18 ± 0.0043
26.72
−0.02 ± 0.0006
0.55
0.00 ± 0.0001
6.61
0.09 ± 0.0008
1.65
−0.29 ± 0.0091
5.76
−0.60 ± 0.0068
30.61
−0.28 ± 0.0044
27.58
0.00 ± 0.0039
12.42
0.07 ± 0.0014
21.52
0.07 ± 0.0010
1.21
−0.03 ± 0.0007
1.21
to different precipitation patterns. Using soil moisture from another day also changes precipitation. But those changes do not
exceed the model uncertainty in more than 10% of all values. Accordingly, slight and realistic changes in soil moisture lead to
changes in precipitation not larger than changes that can also be caused by choosing a slightly different model set-up.
5
5
Systematics
After determining the significance of the strength of precipitation changes, this section handles the systematics of changes.
Significant changes do not necessarily imply systematic changes. While in DRYa (Fig. 6b) predominantly negative amplitude
changes occur, in MOISTp (Fig. 6c) significant but random changes occur. Structure and amplitude change in both positive
and negative directions. To carve out any systematic effects the averaged value of amplitude and structure are compared to the
average of the uncertainty-ensemble. Systematics in L are not analysed as this quantity provides no direction. As explained in
10
Sect. 3 the sample for the SAL results for the uncertainty-ensemble is symmetric and therefore the average is zero. A significant
12
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
difference of the averaged values from zero hints at the systematics. Whether the averaged values differ significantly from zero
is tested statistically by:
zbsys =
5
x1 − x2 − E[x1 − x2 ]
p
.
σ̂ 2 (x1 − x2 )
(9)
x1 and xs denotes the averaged values of S or A for the two compared simulations, E[x1 − x2 ] is the expected value for the
differences between the two simulations and is expected to be zero for the null hypothesis, σ̂ 2 (x1 − x2 ) is the variance of
averages.
Only two simulations with overall modified soil moisture have a systematic effect in precipitation structure (table 3). A positive deviation of structure implying less convection is found in the case with reduced soil moisture for the analysed area “red”,
whereas a negative deviation is found in a case with enhanced soil moisture in region “blue”. Precipitation’s amplitude reacts
10
more often systematically in the analysis for both regions. Modifications of soil moisture by increasing and decreasing result
both in reduced precipitation rates. This implies negative and positive feedback respectively. The positive feedback by decreasing the soil moisture is in consent with Barthlott and Kalthoff (2011). The case study from Barthlott and Kalthoff (2011) show
positive feedback for decreased soil moisture. But enhanced soil moisture can lead to increase or decrease in precipitation,
dependent on the percentage of soil moisture enhancement. In contrast Cheng and Cotton (2004); Ek and Holtslag (2004);
15
Martin and Xue (2006); Hohenegger et al. (2009); Weverberg et al. (2010) all found a negative feedback in convection resolving simulations.
The strength of deviation depends on the strength of modification. While a partly increased soil moisture does not lead to
systematic changes the overall enhancement has a systematic effect. The effect of dry soil exceeds the effect of soil moisture
enhancement and shows systematic effects for both implementations. The effects are stronger for overall modifications. Com-
20
paring the results for both regions the averaged differences calculated for region “blue” exceed those of region “red” because
convective cells are more influenced by soil moisture changes.
6
Conclusion and outlook
The selected case study for 03 August 2012 analysed by the SAL-score provide some results on strength and systematic of soil
moisture influence on precipitation:
25
– Intensive soil moisture modification via artificial enhancement and reduction of soil moisture results in significant
changes in precipitation. Large-area modifications show stronger effects than modification in sub domains.
– Unsystematic changes often occur in structure within an ensemble with same soil moisture modification. Systematic
changes occur often in amplitude within an ensemble with same artificial soil moisture modification.
30
– No systematic in amplitude for all different soil moisture modifications exists. Increase as well as decrease will lead to
systematic negative deviations. Changes in structure show too few systematic changes to allocate an all over systematic.
13
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
– No deviations exceeding the model uncertainty arise by redistributing soil moisture in four bands in this case study. For
differences in precipitation’s location a significant change can be determined for analysis in the smaller terrain.
– Precipitation differences between the CTRL run and simulations with realistic soil moisture modification can not be
proofed as caused by the soil moisture modification. That again shows the difficulties to carve out resilient soil moisture
5
influence.
The results of the two analysis areas differ especially in the percentage of differences exceeding the uncertainty. Having a look
on another precipitation quantity or over a different time interval the results will also look a little different. Furthermore, these
results base on a single case study. Further case studies with less precipitation in the CTRL run and different synoptic forcing
might bring some more different results, especially in systematics of precipitation. To proceed this study the results will be
10
compared to high resolved radar data (Lengfeld et al., 2014). With a larger model domain the uncertainty from the boundary
data could be reduced. If soil moisture effects can be better carved out, model simulation with calculated soil moisture from
radar data will show the possibility to improve simulation of convective precipitation.
14
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
References
Adler, B., Kalthoff, N., and Gantner, L.: The impact of soil moisture inhomogeneities on the modification of a mesoscale convective system:
An idealised model study, ATMOSPHERIC RESEARCH, 101, 354–372, doi:10.1016/j.atmosres.2011.03.013, 2011.
Barthlott, C. and Kalthoff, N.: A Numerical Sensitivity Study on the Impact of Soil Moisture on Convection-Related Parameters and
5
Convective Precipitation over Complex Terrain, Journal of the Atmospheric Sciences, 68, 2971–2987, doi:10.1175/JAS-D-11-027.1,
http://dx.doi.org/10.1175/JAS-D-11-027.1, 2011.
Barthlott, C., Hauck, C., Schädler, G., Kalthoff, N., and Kottmeier, C.: Soil moisture impacts on convective indices and precipitation over
complex terrain, Meteorologische Zeitschrift, 20, 185–197, doi:10.1127/0941-2948/2011/0216, 2011.
Cheng, W. Y. Y. and Cotton, W. R.: Sensitivity of a Cloud-Resolving Simulation of the Genesis of a Mesoscale Convective Sys-
10
tem to Horizontal Heterogeneities in Soil Moisture Initialization, Journal of Hydrometeorology, 5, 934–958, doi:10.1175/15257541(2004)005<0934:SOACSO>2.0.CO;2,
http://journals.ametsoc.org/doi/abs/10.1175/1525-7541(2004)005{\T1\textless}0934:
SOACSO{\T1\textgreater}2.0.CO;2, 2004.
Clark, D., Taylor, C., and Thorpe, A.: Feedback between the land surface and rainfall at convective length scales, JOURNAL OF HYDROMETEOROLOGY, 5, 625–639, doi:10.1175/1525-7541(2004)005<0625:FBTLSA>2.0.CO;2, 2004.
15
Dixon, N. S., Parker, D. J., Taylor, C. M., Garcia-Carreras, L., Harris, P. P., Marsham, J. H., Polcher, J., and Woolley, A.: The effect
of background wind on mesoscale circulations above variable soil moisture in the Sahel, QUARTERLY JOURNAL OF THE ROYAL
METEOROLOGICAL SOCIETY, 139, 1009–1024, 2013.
Doms, G., Förstner, J., Heise, E., Herzog, H.-J., Mironov, D., Raschendorfer, M., Reinhardt, T., Ritter, B., Schrodin, R., Schulz, J.-P., and
Vogel, G.: A Description of the Nonhydrostatic Regional COSMO Model, Part 2: Physical Parameterization, Tech. rep., P.O. Box 100465,
20
63004 Offenbach, germany, 2011.
Ek, M. B. and Holtslag, A. A. M.: Influence of Soil Moisture on Boundary Layer Cloud Development, Journal of Hydrometeorology, 5, 86–
99, doi:10.1175/1525-7541(2004)005<0086:IOSMOB>2.0.CO;2, http://journals.ametsoc.org/doi/abs/10.1175/1525-7541(2004)005{\T1\
textless}0086:IOSMOB{\T1\textgreater}2.0.CO;2, 2004.
Field, P. R. and Heymsfield, A. J.: Importance of snow to global precipitation, Geophysical Research Letters, 42, 9512–9520,
25
doi:10.1002/2015GL065497, 2015.
Froidevaux, P., Schlemmer, L., Schmidli, J., Langhans, W., and Schaer, C.: Influence of the Background Wind on the Local Soil MoisturePrecipitation Feedback, JOURNAL OF THE ATMOSPHERIC SCIENCES, 71, 782–799, doi:10.1175/JAS-D-13-0180.1, 2014.
Glassmeier, F. and Lohmann, U.: Constraining precipitation susceptibility of warm, ice- and mixed-phase clouds with microphysical equations, Journal of the Atmospheric Sciences, pp. JAS–D–16–0008.1, doi:10.1175/JAS-D-16-0008.1, http://journals.ametsoc.org/doi/10.
30
1175/JAS-D-16-0008.1, 2016.
Hauck, C., Barthlott, C., Krauss, L., and Kalthoff, N.: Soil moisture variability and its influence on convective precipitation over complex
terrain, QUARTERLY JOURNAL OF THE ROYAL METEOROLOGICAL SOCIETY, 137, 42–56, doi:10.1002/qj.766, 2011.
Hohenegger, C. and Schär, C.: Predictability and error growth dynamics in cloud-resolving models, JOURNAL OF THE ATMOSPHERIC
SCIENCES, 64, 4467–4478, doi:10.1175/2007JAS2143.1, 2007.
35
Hohenegger, C., Brockhaus, P., Bretherton, C. S., and Schär, C.: The soil moisture-precipitation feedback in simulations with explicit and
parameterized convection, Journal of Climate, 22, 5003–5020, doi:10.1175/2009JCLI2604.1, 2009.
15
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Hohenegger, C., Brockhaus, P., and Schaer, C.: Towards climate simulations at cloud-resolving scales, METEOROLOGISCHE
ZEITSCHRIFT, 17, 383–394, doi:10.1127/0941-2948/2008/0303, 2008.
Kalthoff, N., Kohler, M., Barthlott, C., Adler, B., Mobbs, S. D., Corsmeier, U., Traeumner, K., Foken, T., Eigenmann, R., Krauss, L.,
Khodayar, S., and Di Girolamo, P.: The dependence of convection-related parameters on surface and boundary-layer conditions over
5
complex terrain, QUARTERLY JOURNAL OF THE ROYAL METEOROLOGICAL SOCIETY, 137, 70–80, doi:10.1002/qj.686, 2011.
Kang, S.-K. and Bryan, G. H.: A Large-Eddy Simulation Study of Moist Convection Initiation over Heterogeneous Surface Fluxes,
MONTHLY WEATHER REVIEW, 139, 2901–2917, doi:10.1175/MWR-D-10-05037.1, 2011.
Klüpfel, V., Kalthoff, N., Gantner, L., and Kottmeier, C.: Evaluation of soil moisture ensemble runs to estimate precipitation variability in convection-permitting model simulations for West Africa, ATMOSPHERIC RESEARCH, 101, 178–193,
10
doi:10.1016/j.atmosres.2011.02.008, 2011.
Lengfeld, K., Clemens, M., Münster, H., and Ament, F.: Performance of high-resolution X-band weather radar networks - The PATTERN
example, Atmospheric Measurement Techniques, 7, 4151–4166, doi:10.5194/amt-7-4151-2014, 2014.
Leutwyler, D., Fuhrer, O., Lapillonne, X., Lüthi, D., and Schär, C.: Towards {E}uropean-scale convection-resolving climate simulations with
{GPU}s: a study with {COSMO} 4 .19, Geoscientific Model Development, 9, 3393–3412, doi:10.5194/gmd-9-3393-2016, 2016.
15
Maronga, B. and Raasch, S.: Large-Eddy Simulations of Surface Heterogeneity Effects on the Convective Boundary Layer During the
LITFASS-2003 Experiment, BOUNDARY-LAYER METEOROLOGY, 146, 17–44, doi:10.1007/s10546-012-9748-z, 2013.
Martin, W. J. and Xue, M.: Sensitivity analysis of convection of the 24 May 2002 IHOP case using very large ensembles, Monthly Weather
Review, 134, 192–207, doi:10.1007/11768012_21, 2006.
Mass, C. F., Ovens, D., Westrick, K., and Colle, B. A.: Does increasing horizontal resolution produce more skillful forecasts? The results
20
of two years of real-time numerical weather prediction over the Pacific Northwest, Bulletin of the American Meteorological Society, 83,
407–430+341, doi:10.1175/1520-0477(2002)083<0407:DIHRPM>2.3.CO;2, 2002.
Muhlbauer, a., Hashino, T., Xue, L., Teller, A., Lohmann, U., Rasmussen, R. M., Geresdi, I., and Pan, Z.: Intercomparison of aerosolcloud-precipitation interactions in stratiform orographic mixed-phase clouds, Atmospheric Chemistry and Physics, 10, 8173–8196,
doi:10.5194/acp-10-8173-2010, 2010.
25
Pedersen, L., Jensen, N. E., Christensen, L. E., and Madsen, H.: Quantification of the spatial variability of rainfall based on a dense network of
rain gauges, Atmospheric Research, 95, 441–454, doi:10.1016/j.atmosres.2009.11.007, http://dx.doi.org/10.1016/j.atmosres.2009.11.007,
2010.
Richard, E., Buzzi, A., and Zängl, G.: Quantitative precipitation forecasting in the Alps: The advances achieved by the Mesoscale Alpine
Programme, Quarterly Journal of the Royal Meteorological Society, 133, 831–846, doi:10.1002/qj.65, http://doi.wiley.com/10.1002/qj.65,
30
2007.
Schättler, U., G., D., and C., S.: A description of the nonhydrostatic regional COSMO model. Part VII: User’s Guide, 2009.
Wernli, H., Paulat, M., Hagen, M., and Frei, C.: SAL-A Novel Quality Measure for the Verification of Quantitative Precipitation Forecasts,
MONTHLY WEATHER REVIEW, 136, 4470–4487, doi:10.1175/2008MWR2415.1, 2008.
Weverberg, K. V., van Lipzig, N. P. M., Delobbe, L., and Lauwaet, D.: Sensitivity of quantitative precipitation forecast to soil moisture
35
initialization and microphysics parametrization, QUARTERLY JOURNAL OF THE ROYAL METEOROLOGICAL SOCIETY, 136,
978–996, doi:10.1002/qj.611, 2010.
16
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1049, 2017
Manuscript under review for journal Atmos. Chem. Phys.
Discussion started: 13 February 2017
c Author(s) 2017. CC-BY 3.0 License.
Data availability
Model output data are stored at DKRZ computers and can be provided on demand.
Acknowledgements. We would like to thank the COSMO consortium for access to the code, and the German weather service (DWD) for
providing analysis data, Deutsches Klimarechenzentrum (DKRZ) for providing a simulation platform and Heini Wernli and Markus Zimmer
5
for the SAL analysis code.
17